[
https://issues.apache.org/jira/browse/HDFS-8708?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chengbing Liu reopened HDFS-8708:
---------------------------------
I have different opinion so I'm reopening this issue.
In our production environment, we have both HA and non-HA clusters. A client
should be able to access both kinds of clusters. This is our dilemma.
By setting dfs.client.retry.policy.enabled = true, currently we see:
1) HA nameservice: in case of nn1 shutdown, will still attempt connecting to
nn1 many times (11min by default) before failover, which is undesired
2) non-HA namenode: keep retrying to connect for 11min by default
By setting dfs.client.retry.policy.enabled = false, currently we see:
1) HA nameservice: fast failover, everything works fine
2) non-HA namenode: no retry will be made in case of connection failure, which
is undesired
We would like to ensure fast failover with HA mode as well as multiple retries
with non-HA mode, and we cannot achieve this with current implementation.
Proposed code change:
In {{NameNodeProxiesClient.createProxyWithAlignmentContext}}, {{defaultPolicy}}
should not be passed to {{ClientProtocol}} when {{withRetries}} is false (HA
mode). Instead, TRY_ONCE_THEN_FAIL can be used to ensure fast failover.
> DFSClient should ignore dfs.client.retry.policy.enabled for HA proxies
> ----------------------------------------------------------------------
>
> Key: HDFS-8708
> URL: https://issues.apache.org/jira/browse/HDFS-8708
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Jitendra Nath Pandey
> Assignee: Brahma Reddy Battula
> Priority: Critical
>
> DFSClient should ignore dfs.client.retry.policy.enabled for HA proxies to
> ensure fast failover. Otherwise, dfsclient retries the NN which is no longer
> active and delays the failover.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]